US Federal Reserve: reformat wpic

Embed Doc
Copy Link
Readcast
Collections

CommentGo Back

Download

CREATING A HISTORICAL BRIDGE FOR MANUFACTURING BETWEEN THE STANDARDINDUSTRIAL CLASSIFICATION SYSTEM AND THE NORTH AMERICAN INDUSTRYCLASSIFICATION SYSTEMKimberly N. Bayard and Shawn D. Klimek

Board of Governors of the Federal Reserve and the U.S. Census Bureau

I. Introduction

Since the 1930s, the U.S. government hasclassified industries according to the StandardIndustrial Classification (SIC) system. The SICsystem has been periodically updated and revised tomirror changes in the underlying structure of theeconomy; the most recent change was in 1987. Inthe early 1990s, a movement began to construct amore sweeping revision to industry classification.In 1997, representatives from the statisticalcommunities of the U.S., Canada, and Mexicofinalized the new North American IndustryClassification System (NAICS). NAICS wasdesigned to reflect the significant changes in theeconomy in the 1990s and to standardize industrydefinitions across North America.

The Census Bureau was one of the firstU.S. government entities to collect and publish NAICS-based data for the manufacturing sector with the release of the 1997 Census of Manufactures(COM). In addition, the Census Bureau publishedSIC-based statistics for the 1997 COM, and alsoissued a correspondence between the twoclassification systems. However, the Census Bureaudid not release NAICS-based manufacturingstatistics for any year prior to 1997.This paper documents a joint FederalReserve Board-Census Bureau project to convert the plant-level information in the Censuses of Manufactures from the SIC system to NAICS, anddescribes the subsequent historical SIC-NAICSconcordance that arises. Economists at the FederalReserve Board used this concordance to obtain, back to 1972, NAICS-based output benchmarks.These benchmarks were used in the construction of the recently published NAICS-based historicalrevision of the index of Industrial Production (IP)and the related measures of capacity and capacityutilization.The issuance of a NAICS-based industrytime series for IP fills a gap for users of industrystatistics. The statistical community’s transition to NAICS created a break in industry-level data thathas made it difficult at best to perform basicmanipulations such as trend analysis, forecasting,and seasonal adjustment. The Federal ReserveBoard’s issuance of historical NAICS-based timeseries for IP provides data users with many of thetools necessary to perform standard analyses.

II. Motivation and Data Source

One of the most useful data products provided by the Census Bureau following theissuance of the 1997 COM industry statistics (and,indeed all Economic Censuses) is the official bridgetable between SIC and NAICS. This bridge provides information for each four-digit SICindustry that allows one to calculate the share of establishments, shipments, employment, and payrollassociated with its corresponding six-digit NAICSindustries. Table 1 shows an example of the type of information conveyed in the bridge table for SIC3578, (Calculating & accounting machines, exceptelectronic computers), and how this industrycorresponds to its two associated NAICS industries(Office machinery manufacturing and other computer peripheral equipment manufacturing).

Table 1: Sample bridge between SIC and NAICS for SIC 3578 (1997)

NAICSValue of Shipmts($1,000)Shr of Ship.PaidEmplShr of Empl.Calc. & acc.machines, ex.comp.2,014,806 -- 7,683 --333313Ofc mach.mfg144,380 0.07 966 0.13334119Other comp. peripheralequip. mfg1,870,426 0.93 6,717 0.87

From the table above, it is clear that, in1997, most of SIC 3578 corresponded to “other computer peripheral equipment manufacturing”(NAICS 334119), which includes products such ascomputer monitors, keyboards, printers, andautomatic teller machines. A much smaller share of SIC 3578 corresponded to “office machinerymanufacturing” (NAICS 333313), which includes

A history and description of the various industryclassification systems can be found in the manuals(Executive Office of the President, 1998, 1987,1972).

Some industry concordances are omitted for failureto meet disclosure requirements. An electronic copyof the complete bridge table can be found athttp://www.census.gov/epcd/ec97brdg.

the production of mailhandling machinery andequipment, calculators, and typewriters.The table also illustrates why it issomewhat limiting that the only publicly availableSIC-NAICS bridge that exists is for 1997. Onemight expect that the value of shipments of typewriter and calculator manufacturing relative tocomputer monitors and printers was much greater in, say, 1972 than it was in 1997. However, becausethere is no available information for prior years, datausers can only speculate as to how to construct therelevant breakdown.The absence of historical industry data thatallowed for the construction of a bridge betweenSIC system and NAICS led us to construct such a bridge directly from plant-level data. The Center for Economic Studies at the Census Bureau maintainsthe Longitudinal Research Database (LRD), adatabase of plant-level records for all establishmentsthat appeared in either a COM or an ASM since1963. We used these microdata files to convert theindustry code for each establishment in each COMfrom the SIC basis to NAICS.Specifically, we rely on four informationsources to convert the censuses to a NAICS basis:the Census Bureau’s product level concordancetable that shows a translation of 1987 SIC productcodes to 1992 SIC product codes to 1997 NAICS product codes; the 1997 census that contains bothSIC and NAICS codes for each establishment; plant-specific identifiers that do not change over time, andthe existence of both current-year and the 1987 SICequivalent industry codes for all establishments oncensus files prior to 1997. These last two pieces of information result from CES-specific work tostandardize over time all files in the LRD. As for the industry information, because all LRDestablishments have 1987 SIC information, thetranslation of SIC to NAICS is possible even when asingle NAICS code corresponds to multiple SICcodes.

III. Methodology

Our general approach to assignestablishments to NAICS industries takes place inreverse chronological order. The 1997 COMcontains both SIC and NAICS industry codes and provides our initial link between SIC and NAICSclassifications. As described in more detail below,we use the 1997 classifications as a base to helprecode the 1992 COM to NAICS. Once allestablishments in the 1992 COM are reclassified to NAICS, these industry codes are then used to helpin the reclassification of the 1987 COM, and this process is repeated back to the 1963 COM.

Within each COM year, we assignestablishments to NAICS industries in a series of successive steps. Because the LRD contains a lot of information for each establishment, there are oftenmany options for how we wish to go about theindustry assignment process. We believe, however,that some methods are likely to be more reliablethan others. We first attempt to assign eachestablishment to a NAICS industry by the mostreliable method; any establishment that can’t bematched is considered to be a residual and weattempt to match it in a subsequent step.Briefly, we assign establishments toindustries in one of two general ways: exactmatching or statistical matching. The exactmatching procedures include a “product method”where we classify the types of products made by the plant, an “industry method” where we classify a plant to a NAICS industry based on its SIC industry,an “establishment method” where we identify theSIC industry of an establishment at a certain point intime and use the plant’s presence and industry codein earlier censuses to assign it to a NAICS industry.The statistical assignment procedure looks at a set of characteristics of each plant to help determinewhether they are more consistent with assignment of that plant to one particular NAICS industry or another. There are a few “special cases,” that is, plants that we cannot assign by one of the methodslisted above, and we resort to either shipments-weighted or manual assignment.

Step 1: Product Method

The first step in reclassifying anestablishment to a NAICS industry uses productlevel information collected by the Census Bureaufor some establishments in the COM. The productcode is essentially an extension of the SIC-basedindustry code where the first few digits are identicalto the industry code and the last several digitsspecify a particular product within that industry.The plant is assigned to the industry thatcorresponds to the products that represent greatestvalue of the establishment’s shipments. A plantmay make several products, some of which may falloutside of the establishment’s assigned industry (see below for an illustration).We use a concordance that maps these SIC product codes to NAICS product codes. Sometimes,we can assign NAICS product codes to all of an

COMs are generally conducted every five years.The LRD contains information from the COMsconducted in 1997, 1992, 1987, 1982, 1977, 1972,1967, and 1963.

establishment’s products; other times we can onlyassign a subset of them. When we assign NAICS product codes to the majority of the establishment’soverall shipments, we then use these product codesas a guide to determine the NAICS industryassociated with the greatest value of shipments for each establishment.

Step 3: Establishment Method

The next method of assigning NAICSindustries to manufacturing establishments usesunchanging establishment identifiers to link plantsacross censuses. Recall that our assignmentapproach takes place in reverse chronological order.Therefore, once we have assigned a NAICS code toan establishment in, say, 1992, we can use thisinformation to help inform the assignment of the plant in 1987. We use the plant identifiers to matchestablishments from the census in year t to thecensus in year t-5, and if the plant operates in thesame four digit SIC in both censuses, then we assignthe NAICS industry for the establishment in year tto the establishment in year t-5. Implicitly, thismethod assumes that there is no change within theestablishment that would cause a change inclassification between censuses.Consider the following hypothetical plantthat produces three products and whose primaryindustry is 2282 (represented by the first four digitsof the product code):Product 1: SIC

=2282231, shipments=4000

NAICS

=313112xxxxProduct 2: SIC

=2282241, shipments=3000

NAICS

=313312xxxxProduct 3: SIC

=2269021, shipments=3000

NAICS

=313312xxxx

Step 4: Multiple Characteristics Method(“Statistical Matching”)

In this case, 70 percent of the plant’s shipments in1992 are in SIC 2282. After matching in NAICS product data, we see that the primary industry for the establishment derived from the product codes is NAICS 313312, which accounts for 60 percent of the establishment output. If we had assigned a NAICS industry to the establishment solely basedon the primary four-digit SIC industry, then the plant would be classified into industry 313112.

The next method of assigningestablishments to NAICS industries uses a statisticalmodel that examines characteristics of each plant tohelp determine whether the plant “looks like” other plants in one particular NAICS industry or another.The assignment techniques described in the firstthree steps work well for some establishments andindustries, but are less useful for classifying those plants that are small so that Census does not collectdata on their products, or are in SIC industries thatmap to multiple NAICS industries, or do not survivefrom census to census. As we illustrate below, the“statistical assignment” group accounts for arelatively small fraction of establishments andemployment.We can also assign a NAICS industry whenonly a subset of SIC products match to a unique NAICS product code. For example, for thehypothetical plant above, even if we had been ableto match only product 2 and product 3 to NAICS products codes, we would have been able to assignthe establishment to the proper industry. Other combinations of products matched to NAICS product codes are insufficient to assign the plant bythis method. For all cases where we cannot assignmore than half of the plant’s shipments to a NAICSindustry, we turn to another method of industryassignment.Recalling that our assignment procedurestarts with the most recent census and works backwards, the basic idea behind our statisticalassignment method is to look at the characteristicsof the plants in the same SIC industry in year t, and, based on the distribution in the subsequent (t+5)census, determine the probability that each plantgets assigned to each of its corresponding NAICSindustries.

Step 2: Industry Method

The next method of assigning NAICSindustries uses an industry-level concordance file.Many four-digit SIC industries correspond to onlyone six-digit NAICS industry; we assign plants inthese SIC industries to the relevant NAICS industry.Only establishments in those industries with a one-to-one correspondence can be assigned by thismethod. More than 80 percent of all SIC industriesin manufacturing are represented by this one-to-onescenario. The remaining SIC industries are split intoas many as seven NAICS industries.Specifically, we run a multinomial logitregression for each SIC industry associated withmore than one NAICS industry using the data for allobservations in the industry in the census of year t+5. We run the following model:

inmsa swted tvsted ted NAICS

e b b b b b

+++++=

)() _ ( ) _ () _ (

54321

The variable

is the total employment at the plant,

tvste

is the total value of shipments per worker,

swte

is the average wage, and

inmsa

is a dummy for whether or not the establishment is in an

The NAICS manual shows that SIC 2282 matchesonly to NAICS 313112.

of 00

Category:	Uncategorized.
Rating:
Upload Date:	01/23/2008
Copyright:	Attribution Non-commercial
Tags:	Science money us finance usa unitedstates federalreserve fed thefed federalreservebank Science money us finance usa unitedstates federalreserve fed thefed federalreservebank (fewer)

Documents

People